Metabolomic Analysis of Cold Acclimation of Arctic Mesorhizobium sp. Strain N33

Arctic Mesorhizobium sp. N33 isolated from nodules of Oxytropis arctobia in Canada’s eastern Arctic has a growth temperature range from 0°C to 30°C and is a well-known cold-adapted rhizobia. The key molecular mechanisms underlying cold adaptation in Arctic rhizobia remains totally unknown. Since the concentration and contents of metabolites are closely related to stress adaptation, we applied GC-MS and NMR to identify and quantify fatty acids and water soluble compounds possibly related to low temperature acclimation in strain N33. Bacterial cells were grown at three different growing temperatures (4°C, 10°C and 21°C). Cells from 21°C were also cold-exposed to 4°C for different times (2, 4, 8, 60 and 240 minutes). We identified that poly-unsaturated linoleic acids 18∶2 (9, 12) & 18∶2 (6, 9) were more abundant in cells growing at 4 or 10°C, than in cells cultivated at 21°C. The mono-unsaturated phospho/neutral fatty acids myristoleic acid 14∶1(11) were the most significantly overexpressed (45-fold) after 1hour of exposure to 4°C. As reported in the literature, these fatty acids play important roles in cold adaptability by supplying cell membrane fluidity, and by providing energy to cells. Analysis of water-soluble compounds revealed that isobutyrate, sarcosine, threonine and valine were more accumulated during exposure to 4°C. These metabolites might play a role in conferring cold acclimation to strain N33 at 4°C, probably by acting as cryoprotectants. Isobutyrate was highly upregulated (19.4-fold) during growth at 4°C, thus suggesting that this compound is a precursor for the cold-regulated fatty acids modification to low temperature adaptation.


Introduction
Bacteria have developed many strategies at the transcriptional and post-transcriptional levels to enhance their abilities to withstand cold temperature stress. Some molecular responses to stress are general while others can be specific [1]. For instance, by using genome, cell physiology and transcriptome analyses, it has been shown that at 215uC, the permafrost bacterium Planococcus halocryophilus strain Or1, specifically regulates genes encoding for: the conversion of saturated fatty acids to unsaturated and branched fatty acids; the remodeling of cytoplasmic membrane and cell envelope features; transport systems, chaperone proteins, accumulation of cryoprotectant compounds, and transcriptional regulation [2]. The specific cold adaptive features in bacteria include global resource efficiency, amino acid substitution in coldactive enzymes and increased substrate transport systems [2,3]. Low temperatures also regulate transcripts encoding for specific enzymes (e.g. catalase, glutathione S-transferase, Mo-molybdopterin cofactor biosynthesis and acetyl-CoA dehydrogenase) potentially involved in oxidative stress response by neutralizing toxic compounds such as reactive oxygen species (ROS) [2,4]. A genome sequence analysis study of the psychrophilic Antarctic bacterium Pseudoalteromonas haloplanktis TAC125, suggested that elimination of the entire metabolic pathways involved in ROS generation, can protect the cell against the accumulation of deleterious dioxygen scavenging [5]. Like in many other coldloving bacteria, the synthesis of lipid desaturases has also been detected in strain TAC125. These enzymes increase membrane fluidity, and protect the cell against dioxygen and detoxify the cells at low temperature. On the other hand, as a general response, housekeeping genes are also regulated by a variety of stressors including low temperature which allows these bacteria to downregulate their metabolism to optimize general cell functions in order to withstand cold stresses. Other general response mechanisms include higher turnover of macromolecules, tighter maintenance of intercellular pH, greater osmotic regulation, motility, stopping biomass production and decreasing the activation energy before a pre-exponential growth phase [6]. Reduction of growth and suppression of the genes involved in translation and ribosomal biogenesis in E. coli have been observed and can be considered as a general response and an important strategy under stress conditions [7]. Transcriptional analysis of the Arctic Mesorhizobium N 33 revealed the down-regulation of many housekeeping genes that encode for the cell envelope and outer membrane biogenesis functions, as well as cell motility and secretion at low temperature (unpublished data).
Studies on cold adaptations of the mesophilic [8][9][10] and psychrophillic bacteria [11,12] indicated that they are widely heterogeneous in their genomes content and encompass broad ranges of complex network strategies to survive at low temperature. Most cold adaptation comprehensive studies were mainly performed by applying high-throughput genome sequencing [4,5,11] and other omics technologies such as proteomics [13,14] and transcriptomics [15]. In E. coli, it has been shown that under different perturbations (cold, heat, lactose diauxie, and oxidative stress) the metabolite profiles are much more stressspecific when compared with transcriptomic changes [16].
In our studies, the transcriptional analysis of Arctic Mesorhizobium N 33 at different cold temperature treatments has been informative (unpublished data). For instance, we observed that some amino acids, polyunsaturated fatty acids, and cryoprotectants changed significantly. Since transcriptomics studies only allow the evaluation of molecular adaptation mechanisms at the level of gene expression and did not reflect much higher specificity during early stress adaptation [16], we hypothesised that using metabolomics measurements will allow a better understanding of N 33 cells response to its environmental changes [16][17][18]. Furthermore, metabolites are the end products of cellular processes, and they can be considered as a link between genotype (gene function) and phenotype [19]. It was previously shown that metabolic profiles could be good and specific indicators for monitoring cellular responses to biotic and abiotic stresses [18]. Nevertheless, our understanding of the metabolite content of cold adapted bacteria is very limited.
The psychrotrophic Arctic Mesorhizobium sp. N 33 is the best known cold-adapted symbiotic N 2 -fixing rhizobium isolated from nodules of Oxytropis arctobia, an indigenous legume from Canada's eastern Arctic [20]. Strain N 33 has an intrinsic resistance to streptomycin and it has been conformable to genetic manipulations [21][22][23]. Strain N 33 , like its closely related strain N 31 , has a growth temperature range between 0uC and 30uC, and can establish an efficient symbiosis with the temperate forage crop sainfoin (Onobrychis viciifolia), forming nodules with nitrogenase enzyme (catalyzing the reduction of N 2 to NH 3 + ) active at 10uC [24]. This feature allows legumes to grow in soil that is cold and poor in nitrogen. Since as described earlier, bacteria respond to cold temperature stresses by modifying the level of expression of many genes influencing important cell functions, we hypothesized that in the arctic strain N33, the metabolites found in cell cultures or exposed to suboptimal temperatures, will also vary accordingly. Understanding the variation in N 33 metabolites, will be helpful for future trials aiming at the elucidation of the previously observed cold adaptation of the nitrogenase activity in this Arctic bacterium [24].
Metabolites are very diverse in terms of their chemical structures and abundance [25]. Metabolomic analyses can be performed by using a variety of analytical tools [25][26][27][28]. In this study we applied both nuclear magnetic resonance (NMR) spectroscopy [29] and gas chromatography-mass spectrometry (GC-MS) techniques [25,30] to determine and quantify the water soluble and lipid soluble metabolites [31] of the Arctic Mesorhizobium N 33 subjected to suboptimal temperatures. These data were compared to cells grown at 21uC. Multivariate and univariate statistical analyses, and hierarchical clustering including heatmaps were applied to visualize, identify and compare compounds that were most strongly associated with low temperature adaptation. Our results showed that quantitative metabolomics profiling offers new insights into the chemical environment and metabolic adaptations used by the Arctic Mesorhizobium N 33 at low temperature.

Experimental Design and Metabolic Profiles
GC-MS and NMR were used for the identification and quantification of water soluble metabolites and fatty acids in cells of Mesorhizobium N 33 growing at or exposed to, suboptimal temperatures as compared to cells grown at 21uC considered as control in this study. Water-soluble metabolites and lipids were extracted by phase separation using a biphasic system as described in the materials and methods section. NMR analyses of water soluble metabolites allowed the identification and quantification of 29 compounds ( Table 1, and Table S1). With GC-MS, 26 water-soluble metabolites were identified, of which nine were also detected by NMR ( Table 1 and Figure 1).
Mean concentrations (mM) of water-soluble metabolites (measured by NMR) and their standard deviation (SD) are shown in the supporting information (Table S1). Fatty acids from total, neutral, glyco-and phospho-lipids were identified and quantified using GC-MS. We were able to identify and quantify 13 types of fatty acids from total lipids, 17 from neutral lipids, 17 from phospholipids, and 17 from glycolipids. The complete list of fatty acids, mean concentrations (expressed as mole % of total fatty acids) and their standard deviation (SD) are shown in supporting information (Tables S2, S3, S4, and S5). A list of the 20 fatty acids identified in N 33 is presented in Table 2.

Multivariate Statistical Analyses of Water Soluble Metabolites
Principal components analysis-PCA ( Figure 2A) and heatmap visualization ( Figure S2) show that growth temperature significantly affects water soluble metabolites in Mesorhizobium N33. Cells grown at 10uC had water soluble metabolites contents, comparable to that of the control cells grown at 21uC. Nevertheless, slight differences in the level of metabolites accumulation were observed between the two temperatures. Cells grown at 4uC clustered in a distinct group indicating that at this suboptimal temperature, changes in metabolites levels (up and down) were more pronounced. Out of 29 water soluble metabolites, 19 showed a high accumulation in at least two independent biological repetitions and 10 compounds were suppressed during acclimation to 4uC. Isobutyrate, sarcosine, threonine, and valine were upregulated in all three biological replicates at 4uC ( Figure S2), suggesting that these compounds might be required for energy conservation [2,16]. The significant compounds of N 33 water soluble metabolites after different times of exposure to suboptimal 4uC temperature (T1 to T5; 2 to 240 min), are shown based on PLS-DA (partial least squares discriminate analysis) ( Figure 2B) and heatmap visualization ( Figure S3). The significant (P,0.01) permutation test confirmed the PLS-DA analysis. The optimal PLS-DA model for water soluble metabolites data (measured by NMR) used the top four components with a Q 2 = 0.78. Results revealed that some metabolites accumulated while others decreased after different times of exposure to 4uC compared to the control (T0 = 21uC). The most important compounds are indicated in Figure S2 and they include: sarcosine, glycine, lactate, glutamate, glucose, methionine, 3-hydroxybutyrate, acetate, tyrosine, and isoleucine. The increase in amino-acid levels might result, at least in part, from an increase in protein degradation [32] necessary for the elimination of abnormal proteins resulting from stress conditions. Abundance of the amino-acids might also be due to the demand for the synthesis of new important proteins (chaperones) under stress conditions [16,33]. A genomic study has supported the thermal flexibility apparent from amino acid distribution in the permafrost bacterium (Planococcus halocryophilus strain Or1) [2].

Multivariate Statistical Analyses of Fatty Acids from Neutral Lipids
PCA analysis ( Figure 3A) and heatmap visualization ( Figure  S4) indicated that neutral lipids in cells of Arctic strain N 33 grown at 10uC exhibited fatty acids content comparable to the one observed when cells were grown at 21uC. However a distinct content was obtained when N 33 was cultivated at the suboptimal 4uC temperature. Out of 17 measured fatty acids, linoleic acid 18:2 (6,9) showed higher accumulation in cells grown at 4uC. Many metabolites have shown same change trends at 10 and 21uC ( Figure S4). The significant fatty acids compounds from neutral lipids after different times of exposure to suboptimal 4uC temperature, are shown according to PLS-DA analysis ( Figure 3B), and heatmap visualization ( Figure S5). The permutation test was statistically significant (P,0.07) which confirmed the results of PLS-DA analysis. The optimal PLS-DA model for fatty acid of neutral lipids used the top two components with a Q 2 = 0.37. The PCA analysis ( Figure 4A) of the phospholipids fatty acids content of N 33 growing at constant temperatures and heatmap visualization ( Figure S6) show distinct trend of metabolite changes at each growth temperature tested (21uC, 10uC and 4uC). Out of 17 measured fatty acids from phospholipids, the linoleic acids 18:2(9,12), 18:2 (6,9), palmitoleic acid 16:1(7), and cetoleic acid 22:1(13) showed higher accumulation in N 33 growing at 4uC ( Figure S6). Significant changes in phospholipids fatty acids after different times of exposure to suboptimal 4uC temperature revealed by PLS-DA analysis ( Figure 4B), and heatmap visualization are shown in Figure S7. The significant permutation test (P,0.01) confirms the results of PLS-DA analysis. The optimal PLS-DA model for phospholipids fatty acids used the top five component with a Q 2 = 0.74. The PLS-DA and heatmap show increases or decreases in metabolite after each time of exposure (from T0 to T5) to 4uC.

Multivariate Statistical Analysis of Fatty Acids from Glycolipids
The PCA analysis ( Figure 5A) of the glycolipids fatty acids present in N 33 growing at constant temperatures and heatmap visualization ( Figure S8) show different trends of metabolite changes. N 33 cells growing at 4uC show a distinct trend as compared to cells cultivated at 21uC, whereas an intermediate level of metabolite changes is observed with cells growing at 10uC. Out of 17 measured fatty acids from glycolipids, linoleic acid 18:2 (6,9), nonadecanoic acid (C19), myristic (tetradecanoics) acid (C14) concentration increased at 4uC, and linoleic acid 18:2 (6,9) increased at 10uC. These fatty acids were present at lower concentrations in cells cultivated at 21uC ( Figure S8). The significant fatty acids compounds of glycolipids after different times of exposure to suboptimal 4uC temperature identified by PLS-DA analysis ( Figure 5B) and heatmap visualization ( Figure  S9). The significant permutation test (P,0.01) confirms the results of PLS-DA analysis. The optimal PLS-DA model for fatty acid from glycolipids used the top three component with a Q 2 = 0.75. The PLS-DA and heatmap showed that time of exposure to suboptimal 4uC significantly influenced the concentration of glycolipids fatty acids in the arctic strain N 33 ( Figure 5B).

Multivariate Statistical Analysis of Fatty Acids from Total Lipids
The PCA analysis ( Figure 6A) and heatmap visualization ( Figure S10) show that the total fatty acids content of N 33 is significantly affected by growth temperature. Distinct trends of metabolite changes at 21uC, 10uC and 4uC were observed by heatmap visualization ( Figure S10). Out of 13 total fatty acids, at least 6 compounds showed accumulation at 4uC and 7 others decreased. These patterns were opposite in N 33 growing at 21 and 10uC. Significant fatty acids from total lipids after different times of exposure to suboptimal 4uC temperature revealed by PLS-DA analysis ( Figure 6B), and heatmap visualization is shown in Figure S11. The significant permutation test (P,0.01) confirms the results of the PLS-DA analysis. The optimal PLS-DA model for fatty acids from total lipids used only the top component with a Q 2 of 0.31. The PLS-DA and heatmap showed different trends of metabolite changes for each time of exposure to 4uC.
Unsupervised PCA analysis of fatty acids from the 4 lipid classes, disclosed that the metabolite profiles of N 33 cells grown at 21uC, 4uC or 10uC are each clustered into a distinct group, contrary to cells grown at 21uC and exposed to 4uC for 2 to 240 minutes, thereby suggesting a specific metabolic acclimation to cold.
To distinguish and better understand the detailed information arising from the multivariate analyses, and to identify metabolite features that are significantly different between each treatment and the control, univariate statistical analysis (i.e. t-tests, determining the fold change (FC) and volcano plots) were performed. These analyses were applied to five different metabolite data sets (total fatty acids, and fatty acids from neutral, glyco-and phospholipids, and water soluble compounds) and are presented below.
The maximum increases in fold-change (FC) were observed when N 33 was grown at 4uC or 10uC (GT4 or GT10). In fact, when grown at 4uC and 10uC, N 33 contained respectively 18 and 17 times more linoleic acid 18:2 (6,9) than cells grown at 21uC. The concentration of the fatty acid 19:1(10) significantly decreased in cells grown at 4uC. N 33 cells growing at 21uC (T0) then exposed to 4uC for 2 min (T1) did not exhibit any significant change in fatty acids measured from total lipids. However, when exposed to cold for 4 or 8 min (T2 and T3), 5.6 and 2.2 fold increases were observed for fatty acids 22:1 and 19:1(10) ( Figure 7A).
Univariate statistical analysis indicated that the concentration of 13 (5 saturated and 8 unsaturated) fatty acids from neutral lipids  Figure 7B). Overall, 12 accumulations and 13 reductions in fatty acids from neutral lipids were observed with all cold treatments. As observed with fatty acids of total lipids, at 4uC N 33 cells contained more (FC = 35) fatty acid 18:2 (6,9) than cells of the control (growth at 21uC). Comparable results were observed at 10uC (FC = 6). An important increase (FC = 45) in the concentration of fatty acid 14:1 (11) was observed only in N 33 cells exposed to 4uC for1 h. Under the same condition, the highest fold change ratio (FC = 9) was observed for the saturated fatty acid 19:0. The most important decrease in concentration was observed for the saturated fatty acid 20:0 (FC = 0.27), in N 33 cells after 8 min exposure to 4uC ( Figure 7B). The significant changes in fatty acids from glycolipids of Mesorhizobium N 33 exposed to cold temperature are shown in Figure 7C. Out of the 17 fatty acids detected from glycolipids ( Figure 1 and Table S4), 4 saturated and 6 unsaturated changed concentrations after cold treatments. Significant (P#0.05, FC $2) levels of accumulation resulting from exposure to cold, ranged from 2.32 FC for fatty acid chain 22:1 (13) at 4uC to 12.5 FC for 18:2 (6,9) in N 33 growing at 10uC.
This study shows that some saturated and many unsaturated fatty acids of the Arctic Mesorhizobium N 33 were affected by suboptimal temperatures. Saturated fatty acids have the least steric interference with neighboring methylene groups, and so they may be used to enhance the rigid structure of the membrane and protect the cell conformation. Protection of psychrophilic bacteria from cold temperatures by the overproduction of polyunsaturated fatty acid was previously demonstrated by using transcriptional and biochemical analyses [38,39]. Unsaturated fatty acids have kinks which limit acyl-chain packing and cause a melting point reduction (melting point for linoleic acid 18:2(9, 12) is 28.5uC). This feature improves cell membrane fluidity at low temperatures and the increased production of unsaturated fatty acids appears to be one mechanism for the low temperature adaptation in Arctic Mesorhizobium N 33 . Our results corroborate previous observations indicating that rhizobia increase the production of unsaturated fatty acids during growth at cold temperatures [35,37].
Interestingly, a marked effect of temperature on the fatty acid composition in N 33 was reflected by the different trends observed, when cells were grown at 4uC or 10uC ( Figures 3A to 6A). Nevertheless, at both temperatures, N 33 tends to produce significantly more linoleic acid 18:2 (6,9) and less of 19:1(10) than the control cells for the different types of fatty acids studied. Changes in fatty acid composition following different time of exposure to cold temperatures were more diverse.

Univariate Statistical Analysis of Water Soluble Metabolites
The significant changes (P#0.05, ratio$2) in the water-soluble metabolites content of Mesorhizobium N 33 subjected to different suboptimal temperature treatments were identified by NMR and are summarized in Figure 8. Univariate statistical analysis revealed that the concentration of several compounds were significantly (P#0.05, ratio$2) altered by cold. In comparison to cells grown at 21uC, N 33 grown at 4uC contained significantly more valine, threonine, sarcosine and isobutyrate with FC ranging from 2 to 19.4 ( Figure 8 and Table S1). Seven water soluble compounds showed a significant concentration decrease at 4uC, but the most important were observed with 3-hydroxybutyrate and oxypurinol. Fewer changes were observed in N 33 grown at 10uC compared to 21uC. In fact only 3 compounds changed concentrations and N-carbamoyl-b-alanine displaying the lowest FC ( Figure 8). When grown at 21uC and then exposed to 4uC for different periods of time, in general, N 33 showed more decreases than increases in water soluble metabolites, when compared to the control cells grown at 21uC (Figure 8, Table S1).
The metabolite with the highest fold change for all temperature conditions was isobutyrate, with a 19.4 fold change increase found in N 33 growing at 4uC (GT4). The greatest reduction in water-soluble metabolites was observed at some conditions with a 0.003 fold change. This lowest level of metabolites fold change might be either the result of being below the instrument detection limit or that the level of the metabolite production has been very low compared to control conditions (N 33 grown at 21uC). As a general rule, NMR methods are not particularly sensitive, but using this technique reduces the loss of compounds that may occur during sample preparation [29]. Combining GC-MS and NMR can compensate for the lack of coverage of each platform, but both are still insufficiently sensitive to cover all metabolites. It has been suggested that LC-MS or DI-MS (direct injection of metabolites spectrometry) might be the best methods for metabolomics because of their high sensitivity despite their bias against hydrophilic metabolites [26].
Overall, our metabolomic study of the Arctic Mesorhizobium N 33 using GC-MS and NMR was able to identify 110 compounds involved in central carbon metabolism, essential biosynthetic pathways, secondary metabolism and lipids under different low temperature treatments. GC-MS could measure 64 fatty acids ( Figure 1, Tables S2, S3, S4 and S5) and identify a variety of amino acids and organic acids ( Table 1). NMR spectroscopy provided complementary information by enabling the quantification of 29 water-soluble metabolites ( Figure 1, Table 1).
The metabolomic analysis of Arctic strain N 33 indicated that among the lipid-soluble compounds, poly-unsaturated linoleic acids 18:2(9, 12) and 18:2 (6,9) were the most abundant fatty acids present in cells grown at 4uC or 10uC, as compared to the control cells growing at 21uC. The mono-unsaturated fatty acid (myrestic acid) 14:1(11) from phospho-and neutral lipids was the most significantly overexpressed (45-fold change) after exposure for 60 min to 4uC. These fatty acids are known to provide physical membrane flexibility adaptation and to supply energy to cells [40].
Analysis of water-soluble compounds revealed that isobutyrate, sarcosine, therionine and valine increased during growth at 4uC and after exposure for different times to 4uC, of cells initially grown at 21uC (T0). Among the water-soluble metabolites, isobutyrate was highly upregulated (19.4-fold) in cells grown at 4uC, suggesting that this compound is a precursor for the coldregulated fatty acid modification to low temperature adaptation [41]. We have observed that some metabolites decreased in N 33 under cold conditions. This might be caused by growth cessation or reduction, which is an important strategy to adjust cellular physiology to cold stresses [16].
Sarcosine (N-methylglycine) is an intermediate and a byproduct of glycine synthesis and degradation. Sarcosine oxidase demethylates sarcosine to glycine [42]. It can be derived from the catabolism of betaine, and metabolism of choline, betaine and sarcosine may be linked together [4]. In addition, sarcosine has a significant role in the glutathione metabolic pathway and is a cryoprotectant in psychrophilic bacteria grown at low temperature [5]. These metabolites might have a substantial impact in conferring cold resistance ability to the strain N 33 at 4uC by potentially acting as cryoprotectants. Accumulation of cryoprotectants and the production of unsaturated and short chain fatty acids are considered examples of specific metabolic responses to low temperatures in psychrophillic bacteria as previously shown by transcriptomic and genomic sequencing [4]. Sarcosine is also involved in low temperature osmoadaptation [43] and can be used as a source of carbon, nitrogen and energy [4]. Genome sequencing analysis of the psychrophilic bacterium Colwellia psychrerythraea 34H using genomic and proteomic methodologies [4], and phylogenetic diversity and metabolic potential in a glacier metagenome study revealed that many genes are involved in the synthesis of cryo-osmoprotectants such as glycine, betaine, choline, sarcosine, and glutamate [44]. The cryo-protectants (metabolites and proteins) suppress the aggregation of cellular proteins, stabilize phospholipid bilayers, prevent or reduce ice-crystal formation and freezing damage in bacterial cells at low temperature [45][46][47]. The significantly regulated water and lipid soluble compounds are mainly involved in energy conservation, carbon, protein, nucleic acid, fatty acid, and cryoprotectant biosynthesis. Energy conservation is an essential part of energy stress responses and is associated with many types of stress reactions [5,16].
The metabolite profiles of the Arctic Mesorhizobium N 33 and the changes seen in both metabolite abundance and composition at different cold conditions show that the biochemical changes allowing bacterial cells to tolerate cold is complex. It also suggests that several mechanisms are involved in cold acclimation in the Arctic strain N 33 .
In conclusion, our metabolomic study, using GC-MS and NMR, showed that the Arctic Mesorhizobium N 33 regulates the levels of many compounds, and displays many molecular changes related to cold tolerance.
To identify the pathways and confirm the pattern of important compounds of Arctic bacterium N 33 cold adaptation ability, specific isotope dilution GC-MS or fluxomics techniques are required for accurately quantifying the metabolites in strain N 33 under cold condition. Further investigations combining different omics technologies such as proteomics, genomics, transcriptomics, metabolomics, are required to provide a more complete system biology perspective [18] for a better understanding of the complex mechanisms of cold adaptation in the Arctic Mesorhizobium N 33 .

Strain Cultivation, Experimental Design and Sample Collections
Frozen glycerol stocks of Arctic Mesorhizobium sp. strain N 33 [20] were used to inoculate 20 ml yeast mannitol broth (YMB) medium [48] containing 200 mg ml 21 streptomycin, incubated at 21uC for 5 days on a rotary shaker (180 rpm). The purity of the cells was monitored by plating on solid yeast mannitol agar (YMA) medium containing 200 mg ml 21 streptomycin and 25 mg ml 21 Congo red [49] after 4 days of incubation at 21uC. Subsequent purity test of the strain was performed with a nodulation test on Onobrychis viciifolia (sainfoin) [48] and by 16S-rDNA sequencing [50].
A total of 8 conditions with 3 biological replicates were used in this experiment. After 3 days of growth at 21uC in YMB, 100 mL of pure fresh N 33 cells were used to inoculate 500 ml Erlenmeyer flasks containing 100 ml YMB. Flasks were incubated on a rotary shaker (180 rpm) until cells reached the mid-exponential log phase (OD 600 = 0.4-0.6), under three growing temperature conditions (21uC = control, 4uC, and 10uC; GT21, GT4 and GT10). For the cold stress treatments, mid-log phase cells grown at 21uC (T0) were exposed to 4uC for different times (T1 = 2 min, T2 = 4 min, T3 = 8 min, T4 = 60 min, T5 = 240 min; Figure S1) in a rotary shaker water bath (180 rpm). The samples were immediately transferred into pre-chilled (4uC) Falcon tubes then centrifuged (10,000 g) at 4uC for 5 min. The pellets (100 mg) were washed once with cold TES buffer [51] to eliminate extra-cellular polysaccharides and centrifuged again. The cells were immediately quenched in liquid nitrogen [52] and stored at 285uC. Each 100 ml inoculated YMB provided approximately 100 mg fresh cells. In order to provide enough cell pellets, the experiments were replicated 10 times for each treatment. The pellets from 10 individual flasks per treatment were pooled together before the extraction process and considered as one biological replicate.

Extraction of Lipids and Water-Soluble Metabolites
Bacterial pellets (0.12 g dry weight) were transferred into a 12 ml screw capped glass vials. Lipids were extracted from the bacterial pellets according to the method of Bligh and Dyer [53]. Pellets were homogenized with the Tissue-Tearor (Biospec Products, U.S.A) in 10 ml of 2:1(v/v) chloroform: methanol and shaken thoroughly at 150 rpm for 2 h. The extractant was centrifuged at 3000 rpm for 30 min and the supernatant was transferred into a separate vial. The residue was further extracted with 10 ml of 2:1 (v/v) chloroform: methanol, centrifuged again and both supernatants were pooled. Water soluble metabolites and the lipids were extracted by phase separation of a biphasic system, generated by the addition of one quarter volume of 0.88% potassium chloride solution to the chloroform: methanol extract. This mixture was shaken thoroughly for 30 min and centrifuged. The upper aqueous water-soluble metabolite layer obtained was carefully pipetted into a separate tube and purged with nitrogen gas to remove traces of the solvent. The lower organic lipid layer was evaporated under nitrogen. Lipids were then resuspended in hexane and immediately used or stored under nitrogen at 220uC. The upper aqueous fraction mainly comprised of water soluble metabolites was subsequently frozen in liquid nitrogen and lyophilized.
Total lipid extracts were further separated into neutral lipids, glycolipids and phospholipids classes on LC-Silica Sep Pak cartridges (Supelco) as described by Lynch and Steponkus [54]. Calculated amount of total lipid extract (10-15 mg) dissolved in 1 ml chloroform were transferred to the cartridge. Once the sample had entered the packing, residual sample was washed into the column using 2 ml chloroform followed by an additional 10 to 12 ml of the same solvent to elute neutral lipids. The glycolipids were eluted by the addition of 15 ml acetone: methanol (9:1 v/v) while phospholipids were sequentially eluted using 10 ml methanol. The fractions were dried under nitrogen, resuspended in hexane and immediately used or stored under nitrogen at 220uC.

Preparation of Fatty Acid Methyl Esters (FAMEs)
Fatty acid methyl esters (FAMES) were prepared according to Christie [55]. A known amount of lipids in 2:1 chloroform: methanol (v/v) was mixed with an internal standard (heptadecanoic acid C17:0, Sigma-Aldrich). The mixture was evaporated under nitrogen gas and 1 ml of the methylating reagent (2% sulphuric acid in methanol, v/v) was added. The mixture was incubated at 80uC for 1 h, cooled on ice for 10 min after incubation and neutralized by adding 0.5 ml of a 0.5% sodium chloride solution. FAMEs were extracted twice by vortexing after the addition of 2 ml aliquots of hexane. The two layers were allowed to separate and the upper hexane layer was recovered, and subjected to gas chromatographic analysis for quantification of fatty acids.

Gas Chromatographic Analysis of FAMEs
Gas chromatographic analysis of FAMEs was done using heptadecanoic acid as the internal standard. The analysis was performed on an Agilent 6890N gas chromatography instrument coupled with an Agilent MS-5975 inert XL mass selective detector (Agilent technologies) using the electron impact (EI) ionization mode. Separation of fatty acids was achieved by injecting 2 mL of the FAMEs in a 5% phenyl 95% dimethylpolysiloxane column DB5 (Agilent J & W Scientific, 3060.25 mm60.25 mm). Splitless injection was performed with a constant carrier gas (helium) flow of 1 ml/min. Inlet temperature and transfer line temperatures were set at 250 and 280uC respectively. The temperature programming was as follows: an initial isotherm at 70uC was held for 1 min, raised to 76uC at a rate of 1uC/min, and from 76 to 310uC at a rate of 6.1uC/min. The MS ion source temperature was 230uC and the Quadrupole temperature was 150uC. Peak identification of fatty acids was carried out by comparison of chromatogram with mass spectral library (NIST) and against the retention times and mass spectra of the Supelco 37 component FAME mix (Sigma-Aldrich, St Louis, MO, USA).

GC-MS Method for Identification of Water-Soluble Metabolites
To derivatize the metabolites for GC-MS analysis, 40 mL of methoxyamine hydrochloride (Sigma-Aldrich) in pyridine was added to the water-soluble extracts and incubated at room temperature for 16 hours. Then 50 mL (N-Methyl-N-trifluoroacetamide) MSTFA with 1% TMCS (Trimethylchlorosilane) derivatization agent were added and incubated at 37uC for 60 minutes on a hotplate. The samples were vortexed twice throughout the incubation period to ensure complete dissolution. Samples were refrigerated at 4uC for no longer than 48 hours before analysis in order to avoid any degradation of the derivatized compounds.
Derivatized extracts were analyzed using an Agilent 7890-5975C GC-MS instrument operating in an Electron Impact (EI) ionization mode. For GC-MS analysis, 2 mL of the derivatized samples were injected using a split/splitless injector with a split ratio of 5:1 onto a HP-5MS capillary column (30 m6250 mm60.25 mm). The helium carrier gas was set to a flow rate of 1 mL/min and the initial oven temperature was set to 70uC. The temperature was increased at 1uC/min to 76uC, and then at 6.1uC/min to 310uC. The total run time was 45 minutes. The full scan mode of the quadrupole MS was used at a mass range of 50-500 m/z, with a solvent delay of 6 minutes. The MS ion source temperature was 230uC and the Quadrupole temperature was 150uC. In GC-MS, a faster scan speed generally provides more data points across a chromatographic peak, but it tends to lower the ion statistics. In contrast, a slower scan rate produces few scans over the peak and results in better spectra. The scan speed of our quadrupole MS was optimized over a number of samples and it was found that a relatively slow scan rate of 1.7 scans gave the best results.
The AMDIS spectral deconvolution software (Version 2.62) from NIST (National Institute of Standards and Technology) was used to process the total ion chromatogram and the EI-MS spectra of each GC peak. After deconvolution, the purified mass spectrum of each of the trimethylsilated metabolites was identified using the NIST MS Search program (version 2.0d) which was linked to the NIST mass spectral library (2005). Retention Indices (RIs) were calculated using an external alkane standard. Metabolites were identified not only by matching the EI-MS spectra with the those of reference compounds from NIST library, but also by matching the experimental RI of each metabolite with an in-house RI library (containing 312 TMS-derivatized metabolites) developed in our laboratory [26].

NMR Sample Preparation
To identify and quantify the water-soluble metabolites, the evaporated water soluble fraction from different samples (,42.8 mg/ml extracted cells) was dissolved in 500 mL of 50 mM NaH 2 PO 4 buffer pH 7. Thirty five mL of D 2 O and 15 mL of a buffer solution (0.5 mM DSS (disodium-2,2-dimethyl-2-silapentane-5-sulphonate) and 0.47% NaN 3 in H 2 O) and 1 mM Imidazole were added to the sample. The sample amount in the final assay volume (350 mL) was 15 mg. The sample solution was vortexed for 1 minute, sonicated for 30 minutes, and transferred to a standard Shigemi microcell NMR tube for subsequent spectral analysis.

NMR Spectroscopy
All 1 H-NMR spectra were collected on a 500 MHz Inova (Varian Inc., Palo Alto, CA) spectrometer equipped with a 5 mm HCN Z-gradient pulsed-field gradient (PFG) room-temperature probe. 1 H-NMR spectra were acquired at 25uC using the first transient of the NOESY-presaturation pulse sequence, which was chosen for its high degree of quantitative accuracy. Spectra were collected with 256 transients using a 4s acquisition time and a 1s recycle delay.

NMR Compound Identification and Quantification
All FIDs (free induction decays) were zero-filled to 64k data points and subjected to line broadening of 0.5 Hz. The singlet produced by the DSS methyl groups was used as an internal standard for chemical shift referencing (set to 0 ppm) and for quantification. All 1 H-NMR spectra were processed and analyzed using the Chenomx NMR Suite Professional software package version 6.0 (Chenomx Inc., Edmonton, AB). The Chenomx NMR Suite software allows for qualitative and quantitative analysis of an NMR spectrum by manually fitting spectral signatures from an internal database of reference spectra to the full NMR spectrum [56]. Specifically, the spectral fitting for each metabolite was done using the standard Chenomx 500 MHz metabolite library. Typically 90% of all visible peaks were assigned to a compound and more than 90% of the spectral area could be routinely fit using the Chenomx spectral analysis software. Most of the visible peaks are annotated with a compound name. It has been previously shown that this fitting procedure provides absolute concentration accuracies of 90% or better. Each spectrum was processed and analyzed by at least two NMR spectroscopists to minimize compound misidentification and misquantification. We used sample spiking to confirm the identities of assigned compound. The confirmations of the spiking with standard and original peaks in experiments for those unexpected metabolites (i.e. phenylacetate and N-carbamoyl-beta-alanine) and for those hard to assign (i.e. malonic acid and oxypurinol) are shown in Figure S12 (A-D). Sample spiking involves the addition of 20-200 mM of the suspected compound to selected samples and testing whether the relative NMR signal intensity changed as expected.

Data Processing for Statistical Analysis
To visualize the compounds, and compare the metabolite changes (composition and concentrations) of all cold perturbation treatments with the control (N 33 cells grown at 21uC = GT21 or T0), several multivariate and univariate analytical methods were applied. The water and lipid soluble compound concentration data tables were arranged with samples in column and compounds in rows. Data tables were formatted as comma separated values (.csv). Data tables with three growing temperatures (GT21, GT4, GT10) and 6 time points (T0, T1, T2, T3, T8, T4, T5) were uploaded to the MetaboAnalyst 2.0 server (http://www. metaboanalyst.ca) [57] and analysed separately. In both groups of data metabolite, data were unpaired and analysed using multivariate and univariate methods. To reduce any possible variance and to improve the performance for downstream statistical analysis, metabolites data generated by GC-MS and NMR were normalized using MetaboAnalyst's normalization protocols [57]. For multivariate analysis (PCA and PLS-DA), Row-wise normalization was performed for all metabolite data by comparing the samples with a pooled average of reference samples (cells grown at 21uC) to make each sample comparable to one another. To make the metabolite concentration values more comparable among different compounds, several different types of column-wise normalization were performed. The fractions of fatty acids (from total, neutral, phospho-, and glyco-lipids) were log transformed and analysed individually. The fatty acids data were also normalized using auto scaling (mean-centered and divided by the standard deviation of each variable). Same normalization procedures were performed for NMR data. Univariate analysis was applied to calculate the fold change (FC), volcano plots, and statistical significance (t-test and one way ANOVA) assessed. Since FC is calculated as the ratio between two group means (sample over control), the column-wise normalization (i.e. log transformation/mean-centering) will significantly change the absolute values, thus in univariate analysis data were used before column normalization [58]. To assess the degree of metabolite concentration changes, heat maps and hierarchical clustering were performed using the MetaboAnalyst 2.0 software [57]. Heatmaps were created based on the Pearson distance measure and the Ward clustering algorithm, displayed for top 25 features selected by analysis of variance (ANOVA) using a significance level of P#0.05, and post-hoc analysis of Fisher's LSD. Figure S1 Experimental plan used to study metabolomics of cold adaptation of the Arctic Mesorhizobium N 33 . Bacteria were cultivated in yeast mannitol broth (YMB) medium at a constant temperature of 21uC (GT21), 4uC (GT4) or 10uC (GT10). For the effect of time of exposure to cold, N 33 cells were grown at 21uC (T0) and then exposed to 4uC for: 2 min (T1), 4 min (T2), 8 min (T3), 60 min (T4) and 240 min (T5). Each treatment is color coded, and the color codes are used in all figures of this manuscript. (TIF) Figure S2 Heatmap visualization of water soluble metabolites present in Mesorhizobium N 33 during growth at different temperatures. GT21 = growth at 21uC (control); GT4 = growth at 4uC; GT10 = growth at 10uC. Data were row-wise normalized by a pooled averaged reference sample (GT21), and were auto scaled and log transformed. Hierarchical clustering was performed based on Pearson's distance on 29 water soluble metabolites and is shown at the top and side of the panel. Brown and blue colors represent an increase and decrease of a metabolite. The heatmap visualization shows for each growth temperature used a distinct effect Conditions GT21 and GT10 represent close change trends of the water soluble metabolites. However some compounds have shown slightly different levels of accumulations. Metabolites of the cells grown at 4uC (GT4) are clustered in a distinct group far from those of GT10 and GT21. Most influential compounds that were highly accumulated during constant growth at 4uC include isobutyrate, sarcosine, threonine, and valine. (TIF) Figure S3 Heatmap visualization of water soluble metabolites present in Mesorhizobium N 33 exposed to suboptimal 4uC for various times.  Figure S8 Heatmap visualization of fatty acids from glycolipids present in Mesorhizobium N 33 during growth at different temperatures. GT21 = growth at 21uC (control); GT4 = growth at 4uC; GT10 = growth at 10uC. Data were rowwise normalized by a pooled averaged reference samples (GT21), and were auto scaled and log transformed. Hierarchical clustering was performed based on Pearson's distance on 17 fatty acids from glycolipids and is shown at the top and side of the panel. Brown and blue colors represent an increase and decrease of a metabolite. The heatmap visualization shows different trends of the metabolite changes at 21uC (GT21), 10uC (GT10) and 4uC (GT4). The fatty acids were grouped in 3 main clusters and 5 sub-clusters. Conditions GT4 represents distinct trends of metabolite changes compared to metabolites of the cells grown at 21uC, whereas cells grown at 10uC (GT10) represents an intermediate levels of metabolite changes. (TIF) Figure S9 Heatmap visualization of fatty acids from glycolipids present in Mesorhizobium N 33 exposed to suboptimal 4uC for various times. T0 = 21uC (control), T1 = 2 min; T2 = 4 min; T3 = 8 min; T4 = 60 min; T5 = 240 min exposure to 4uC of cells grown at 21uC. Data were row-wise normalized by a pooled averaged reference sample (T0), and were auto scaled and log transformed. Hierarchical clustering was performed based on Pearson's distance on 17 fatty acids from glycolipids and is shown at the top and side of the panel. Brown and blue colors represent an increase and decrease of a metabolite. The heatmap visualization shows different trends of metabolite changes under each time of exposure to low temperature. (TIF) Figure S10 Heatmap visualization of fatty acids from total lipids present in Mesorhizobium N 33 during growth at different temperatures. GT21 = growth at 21uC (control); GT4 = growth at 4uC; GT10 = growth at 10uC. Data were rowwise normalized by a pooled averaged reference samples (GT21), and were auto scaled and log transformed. Hierarchical clustering was performed based on Pearson's distance on 13 fatty acids and is shown at the top and side of the panel. Brown and blue colors represent an increase and decrease of a metabolite. The heatmap visualization shows distinct trends of the metabolite changes at 21uC (GT21), 10uC (GT10) and 4uC (GT4). The fatty acids were grouped in 3 main clusters and 8 sub-clusters in Heatmaps. Conditions GT4 represents distinct trends of metabolite changes compared to the metabolites of the cells grown at 21uC and 10uC. Out of 13 fatty acids of total lipids, at least 6 fatty acids showed accumulation at 4uC and 7 compounds showed down regulations. (TIF) Figure S11 Heatmap visualization of fatty acids from total lipids present in Mesorhizobium N 33 exposed to suboptimal 4uC for various times.